Fix wrong argument in autoderef process

[ldm0]

The overloaded_deref_ty is a function for derefencing a type which overloads the Deref trait. But actually this function never uses the parameter pushed in until this PR. -_-

[rust-highfive]

r? @petrochenkov

(rust_highfive has picked a reviewer for you, use r? to override)

[Dylan-DPC-zz]

@bors r+ rollup

[bors]

📌 Commit 8d2f301 has been approved by Dylan-DPC

[RalfJung]

Does this change behavior in any way? Is there a test case we could add to ensure it does not regress?

[ldm0]

@RalfJung I'm pretty sure this won't cause a regress. This just correct the function to do the thing it meant to do. Is this related to some bug? I can help reviewing.

[RalfJung]

Sorry, what I meant is -- this fixes a bug, right? If so, there should be a test to make sure that in the future, this does not get changed back to the old, wrong behavior that we had before your PR.

[ldm0]

@RalfJung This doesn't fix a bug directly. This actually fixes future bugs. The situation is like this:

a = 1
b = 2
add(a, b) {
    return a + 2
}
print(add(1, 2))

Here we get correct answer 3. But in the future when we calls add(1, 3), it quietly causes a bug (and it will be extremely hard to digout in such a large codebase). This PR just changes a + 2 to a + b.

And this function is an internal function and only called in one place so for a long time this bug is left undiscovered. I think for regression avoiding, a unit test is needed, but unit tests seem to be not exist in current compiler. :-(